drivers/net/hyperv/netvsc.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <[email protected]>
   7  *   Hank Janssen  <[email protected]>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/kernel.h>
  12 #include <linux/sched.h>
  13 #include <linux/wait.h>
  14 #include <linux/mm.h>
  15 #include <linux/delay.h>
  16 #include <linux/io.h>
  17 #include <linux/slab.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/prefetch.h>
  23
  24 #include <asm/sync_bitops.h>
  25
  26 #include "hyperv_net.h"
  27 #include "netvsc_trace.h"
  28
  29 /*
  30  * Switch the data path from the synthetic interface to the VF
  31  * interface.
  32  */
  33 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  34 {
  35         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  36         struct hv_device *dev = net_device_ctx->device_ctx;
  37         struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
  38         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  39
  40         memset(init_pkt, 0, sizeof(struct nvsp_message));
  41         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  42         if (vf)
  43                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  44                         NVSP_DATAPATH_VF;
  45         else
  46                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  47                         NVSP_DATAPATH_SYNTHETIC;
  48
  49         trace_nvsp_send(ndev, init_pkt);
  50
  51         vmbus_sendpacket(dev->channel, init_pkt,
  52                                sizeof(struct nvsp_message),
  53                                (unsigned long)init_pkt,
  54                                VM_PKT_DATA_INBAND, 0);
  55 }
  56
  57 /* Worker to setup sub channels on initial setup
  58  * Initial hotplug event occurs in softirq context
  59  * and can't wait for channels.
  60  */
  61 static void netvsc_subchan_work(struct work_struct *w)
  62 {
  63         struct netvsc_device *nvdev =
  64                 container_of(w, struct netvsc_device, subchan_work);
  65         struct rndis_device *rdev;
  66         int i, ret;
  67
  68         /* Avoid deadlock with device removal already under RTNL */
  69         if (!rtnl_trylock()) {
  70                 schedule_work(w);
  71                 return;
  72         }
  73
  74         rdev = nvdev->extension;
  75         if (rdev) {
  76                 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
  77                 if (ret == 0) {
  78                         netif_device_attach(rdev->ndev);
  79                 } else {
  80                         /* fallback to only primary channel */
  81                         for (i = 1; i < nvdev->num_chn; i++)
  82                                 netif_napi_del(&nvdev->chan_table[i].napi);
  83
  84                         nvdev->max_chn = 1;
  85                         nvdev->num_chn = 1;
  86                 }
  87         }
  88
  89         rtnl_unlock();
  90 }
  91
  92 static struct netvsc_device *alloc_net_device(void)
  93 {
  94         struct netvsc_device *net_device;
  95
  96         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  97         if (!net_device)
  98                 return NULL;
  99
 100         init_waitqueue_head(&net_device->wait_drain);
 101         net_device->destroy = false;
 102         net_device->tx_disable = true;
 103
 104         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
 105         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
 106
 107         init_completion(&net_device->channel_init_wait);
 108         init_waitqueue_head(&net_device->subchan_open);
 109         INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
 110
 111         return net_device;
 112 }
 113
 114 static void free_netvsc_device(struct rcu_head *head)
 115 {
 116         struct netvsc_device *nvdev
 117                 = container_of(head, struct netvsc_device, rcu);
 118         int i;
 119
 120         kfree(nvdev->extension);
 121         vfree(nvdev->recv_buf);
 122         vfree(nvdev->send_buf);
 123         kfree(nvdev->send_section_map);
 124
 125         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
 126                 xdp_rxq_info_unreg(&nvdev->chan_table[i].xdp_rxq);
 127                 vfree(nvdev->chan_table[i].mrc.slots);
 128         }
 129
 130         kfree(nvdev);
 131 }
 132
 133 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 134 {
 135         call_rcu(&nvdev->rcu, free_netvsc_device);
 136 }
 137
 138 static void netvsc_revoke_recv_buf(struct hv_device *device,
 139                                    struct netvsc_device *net_device,
 140                                    struct net_device *ndev)
 141 {
 142         struct nvsp_message *revoke_packet;
 143         int ret;
 144
 145         /*
 146          * If we got a section count, it means we received a
 147          * SendReceiveBufferComplete msg (ie sent
 148          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 149          * to send a revoke msg here
 150          */
 151         if (net_device->recv_section_cnt) {
 152                 /* Send the revoke receive buffer */
 153                 revoke_packet = &net_device->revoke_packet;
 154                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 155
 156                 revoke_packet->hdr.msg_type =
 157                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 158                 revoke_packet->msg.v1_msg.
 159                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 160
 161                 trace_nvsp_send(ndev, revoke_packet);
 162
 163                 ret = vmbus_sendpacket(device->channel,
 164                                        revoke_packet,
 165                                        sizeof(struct nvsp_message),
 166                                        (unsigned long)revoke_packet,
 167                                        VM_PKT_DATA_INBAND, 0);
 168                 /* If the failure is because the channel is rescinded;
 169                  * ignore the failure since we cannot send on a rescinded
 170                  * channel. This would allow us to properly cleanup
 171                  * even when the channel is rescinded.
 172                  */
 173                 if (device->channel->rescind)
 174                         ret = 0;
 175                 /*
 176                  * If we failed here, we might as well return and
 177                  * have a leak rather than continue and a bugchk
 178                  */
 179                 if (ret != 0) {
 180                         netdev_err(ndev, "unable to send "
 181                                 "revoke receive buffer to netvsp\n");
 182                         return;
 183                 }
 184                 net_device->recv_section_cnt = 0;
 185         }
 186 }
 187
 188 static void netvsc_revoke_send_buf(struct hv_device *device,
 189                                    struct netvsc_device *net_device,
 190                                    struct net_device *ndev)
 191 {
 192         struct nvsp_message *revoke_packet;
 193         int ret;
 194
 195         /* Deal with the send buffer we may have setup.
 196          * If we got a  send section size, it means we received a
 197          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 198          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 199          * to send a revoke msg here
 200          */
 201         if (net_device->send_section_cnt) {
 202                 /* Send the revoke receive buffer */
 203                 revoke_packet = &net_device->revoke_packet;
 204                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 205
 206                 revoke_packet->hdr.msg_type =
 207                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 208                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 209                         NETVSC_SEND_BUFFER_ID;
 210
 211                 trace_nvsp_send(ndev, revoke_packet);
 212
 213                 ret = vmbus_sendpacket(device->channel,
 214                                        revoke_packet,
 215                                        sizeof(struct nvsp_message),
 216                                        (unsigned long)revoke_packet,
 217                                        VM_PKT_DATA_INBAND, 0);
 218
 219                 /* If the failure is because the channel is rescinded;
 220                  * ignore the failure since we cannot send on a rescinded
 221                  * channel. This would allow us to properly cleanup
 222                  * even when the channel is rescinded.
 223                  */
 224                 if (device->channel->rescind)
 225                         ret = 0;
 226
 227                 /* If we failed here, we might as well return and
 228                  * have a leak rather than continue and a bugchk
 229                  */
 230                 if (ret != 0) {
 231                         netdev_err(ndev, "unable to send "
 232                                    "revoke send buffer to netvsp\n");
 233                         return;
 234                 }
 235                 net_device->send_section_cnt = 0;
 236         }
 237 }
 238
 239 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
 240                                        struct netvsc_device *net_device,
 241                                        struct net_device *ndev)
 242 {
 243         int ret;
 244
 245         if (net_device->recv_buf_gpadl_handle) {
 246                 ret = vmbus_teardown_gpadl(device->channel,
 247                                            net_device->recv_buf_gpadl_handle);
 248
 249                 /* If we failed here, we might as well return and have a leak
 250                  * rather than continue and a bugchk
 251                  */
 252                 if (ret != 0) {
 253                         netdev_err(ndev,
 254                                    "unable to teardown receive buffer's gpadl\n");
 255                         return;
 256                 }
 257                 net_device->recv_buf_gpadl_handle = 0;
 258         }
 259 }
 260
 261 static void netvsc_teardown_send_gpadl(struct hv_device *device,
 262                                        struct netvsc_device *net_device,
 263                                        struct net_device *ndev)
 264 {
 265         int ret;
 266
 267         if (net_device->send_buf_gpadl_handle) {
 268                 ret = vmbus_teardown_gpadl(device->channel,
 269                                            net_device->send_buf_gpadl_handle);
 270
 271                 /* If we failed here, we might as well return and have a leak
 272                  * rather than continue and a bugchk
 273                  */
 274                 if (ret != 0) {
 275                         netdev_err(ndev,
 276                                    "unable to teardown send buffer's gpadl\n");
 277                         return;
 278                 }
 279                 net_device->send_buf_gpadl_handle = 0;
 280         }
 281 }
 282
 283 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 284 {
 285         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 286         int node = cpu_to_node(nvchan->channel->target_cpu);
 287         size_t size;
 288
 289         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
 290         nvchan->mrc.slots = vzalloc_node(size, node);
 291         if (!nvchan->mrc.slots)
 292                 nvchan->mrc.slots = vzalloc(size);
 293
 294         return nvchan->mrc.slots ? 0 : -ENOMEM;
 295 }
 296
 297 static int netvsc_init_buf(struct hv_device *device,
 298                            struct netvsc_device *net_device,
 299                            const struct netvsc_device_info *device_info)
 300 {
 301         struct nvsp_1_message_send_receive_buffer_complete *resp;
 302         struct net_device *ndev = hv_get_drvdata(device);
 303         struct nvsp_message *init_packet;
 304         unsigned int buf_size;
 305         size_t map_words;
 306         int ret = 0;
 307
 308         /* Get receive buffer area. */
 309         buf_size = device_info->recv_sections * device_info->recv_section_size;
 310         buf_size = roundup(buf_size, PAGE_SIZE);
 311
 312         /* Legacy hosts only allow smaller receive buffer */
 313         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 314                 buf_size = min_t(unsigned int, buf_size,
 315                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
 316
 317         net_device->recv_buf = vzalloc(buf_size);
 318         if (!net_device->recv_buf) {
 319                 netdev_err(ndev,
 320                            "unable to allocate receive buffer of size %u\n",
 321                            buf_size);
 322                 ret = -ENOMEM;
 323                 goto cleanup;
 324         }
 325
 326         net_device->recv_buf_size = buf_size;
 327
 328         /*
 329          * Establish the gpadl handle for this buffer on this
 330          * channel.  Note: This call uses the vmbus connection rather
 331          * than the channel to establish the gpadl handle.
 332          */
 333         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 334                                     buf_size,
 335                                     &net_device->recv_buf_gpadl_handle);
 336         if (ret != 0) {
 337                 netdev_err(ndev,
 338                         "unable to establish receive buffer's gpadl\n");
 339                 goto cleanup;
 340         }
 341
 342         /* Notify the NetVsp of the gpadl handle */
 343         init_packet = &net_device->channel_init_pkt;
 344         memset(init_packet, 0, sizeof(struct nvsp_message));
 345         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 346         init_packet->msg.v1_msg.send_recv_buf.
 347                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 348         init_packet->msg.v1_msg.
 349                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 350
 351         trace_nvsp_send(ndev, init_packet);
 352
 353         /* Send the gpadl notification request */
 354         ret = vmbus_sendpacket(device->channel, init_packet,
 355                                sizeof(struct nvsp_message),
 356                                (unsigned long)init_packet,
 357                                VM_PKT_DATA_INBAND,
 358                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 359         if (ret != 0) {
 360                 netdev_err(ndev,
 361                         "unable to send receive buffer's gpadl to netvsp\n");
 362                 goto cleanup;
 363         }
 364
 365         wait_for_completion(&net_device->channel_init_wait);
 366
 367         /* Check the response */
 368         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
 369         if (resp->status != NVSP_STAT_SUCCESS) {
 370                 netdev_err(ndev,
 371                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
 372                            resp->status);
 373                 ret = -EINVAL;
 374                 goto cleanup;
 375         }
 376
 377         /* Parse the response */
 378         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
 379                    resp->num_sections, resp->sections[0].sub_alloc_size,
 380                    resp->sections[0].num_sub_allocs);
 381
 382         /* There should only be one section for the entire receive buffer */
 383         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 384                 ret = -EINVAL;
 385                 goto cleanup;
 386         }
 387
 388         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
 389         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 390
 391         /* Setup receive completion ring.
 392          * Add 1 to the recv_section_cnt because at least one entry in a
 393          * ring buffer has to be empty.
 394          */
 395         net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
 396         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 397         if (ret)
 398                 goto cleanup;
 399
 400         /* Now setup the send buffer. */
 401         buf_size = device_info->send_sections * device_info->send_section_size;
 402         buf_size = round_up(buf_size, PAGE_SIZE);
 403
 404         net_device->send_buf = vzalloc(buf_size);
 405         if (!net_device->send_buf) {
 406                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
 407                            buf_size);
 408                 ret = -ENOMEM;
 409                 goto cleanup;
 410         }
 411
 412         /* Establish the gpadl handle for this buffer on this
 413          * channel.  Note: This call uses the vmbus connection rather
 414          * than the channel to establish the gpadl handle.
 415          */
 416         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 417                                     buf_size,
 418                                     &net_device->send_buf_gpadl_handle);
 419         if (ret != 0) {
 420                 netdev_err(ndev,
 421                            "unable to establish send buffer's gpadl\n");
 422                 goto cleanup;
 423         }
 424
 425         /* Notify the NetVsp of the gpadl handle */
 426         init_packet = &net_device->channel_init_pkt;
 427         memset(init_packet, 0, sizeof(struct nvsp_message));
 428         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 429         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 430                 net_device->send_buf_gpadl_handle;
 431         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 432
 433         trace_nvsp_send(ndev, init_packet);
 434
 435         /* Send the gpadl notification request */
 436         ret = vmbus_sendpacket(device->channel, init_packet,
 437                                sizeof(struct nvsp_message),
 438                                (unsigned long)init_packet,
 439                                VM_PKT_DATA_INBAND,
 440                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 441         if (ret != 0) {
 442                 netdev_err(ndev,
 443                            "unable to send send buffer's gpadl to netvsp\n");
 444                 goto cleanup;
 445         }
 446
 447         wait_for_completion(&net_device->channel_init_wait);
 448
 449         /* Check the response */
 450         if (init_packet->msg.v1_msg.
 451             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 452                 netdev_err(ndev, "Unable to complete send buffer "
 453                            "initialization with NetVsp - status %d\n",
 454                            init_packet->msg.v1_msg.
 455                            send_send_buf_complete.status);
 456                 ret = -EINVAL;
 457                 goto cleanup;
 458         }
 459
 460         /* Parse the response */
 461         net_device->send_section_size = init_packet->msg.
 462                                 v1_msg.send_send_buf_complete.section_size;
 463
 464         /* Section count is simply the size divided by the section size. */
 465         net_device->send_section_cnt = buf_size / net_device->send_section_size;
 466
 467         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 468                    net_device->send_section_size, net_device->send_section_cnt);
 469
 470         /* Setup state for managing the send buffer. */
 471         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 472
 473         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 474         if (net_device->send_section_map == NULL) {
 475                 ret = -ENOMEM;
 476                 goto cleanup;
 477         }
 478
 479         goto exit;
 480
 481 cleanup:
 482         netvsc_revoke_recv_buf(device, net_device, ndev);
 483         netvsc_revoke_send_buf(device, net_device, ndev);
 484         netvsc_teardown_recv_gpadl(device, net_device, ndev);
 485         netvsc_teardown_send_gpadl(device, net_device, ndev);
 486
 487 exit:
 488         return ret;
 489 }
 490
 491 /* Negotiate NVSP protocol version */
 492 static int negotiate_nvsp_ver(struct hv_device *device,
 493                               struct netvsc_device *net_device,
 494                               struct nvsp_message *init_packet,
 495                               u32 nvsp_ver)
 496 {
 497         struct net_device *ndev = hv_get_drvdata(device);
 498         int ret;
 499
 500         memset(init_packet, 0, sizeof(struct nvsp_message));
 501         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 502         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 503         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 504         trace_nvsp_send(ndev, init_packet);
 505
 506         /* Send the init request */
 507         ret = vmbus_sendpacket(device->channel, init_packet,
 508                                sizeof(struct nvsp_message),
 509                                (unsigned long)init_packet,
 510                                VM_PKT_DATA_INBAND,
 511                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 512
 513         if (ret != 0)
 514                 return ret;
 515
 516         wait_for_completion(&net_device->channel_init_wait);
 517
 518         if (init_packet->msg.init_msg.init_complete.status !=
 519             NVSP_STAT_SUCCESS)
 520                 return -EINVAL;
 521
 522         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 523                 return 0;
 524
 525         /* NVSPv2 or later: Send NDIS config */
 526         memset(init_packet, 0, sizeof(struct nvsp_message));
 527         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 528         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 529         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 530
 531         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 532                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 533
 534                 /* Teaming bit is needed to receive link speed updates */
 535                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 536         }
 537
 538         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
 539                 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
 540
 541         trace_nvsp_send(ndev, init_packet);
 542
 543         ret = vmbus_sendpacket(device->channel, init_packet,
 544                                 sizeof(struct nvsp_message),
 545                                 (unsigned long)init_packet,
 546                                 VM_PKT_DATA_INBAND, 0);
 547
 548         return ret;
 549 }
 550
 551 static int netvsc_connect_vsp(struct hv_device *device,
 552                               struct netvsc_device *net_device,
 553                               const struct netvsc_device_info *device_info)
 554 {
 555         struct net_device *ndev = hv_get_drvdata(device);
 556         static const u32 ver_list[] = {
 557                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 558                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
 559                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
 560         };
 561         struct nvsp_message *init_packet;
 562         int ndis_version, i, ret;
 563
 564         init_packet = &net_device->channel_init_pkt;
 565
 566         /* Negotiate the latest NVSP protocol supported */
 567         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 568                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 569                                        ver_list[i])  == 0) {
 570                         net_device->nvsp_version = ver_list[i];
 571                         break;
 572                 }
 573
 574         if (i < 0) {
 575                 ret = -EPROTO;
 576                 goto cleanup;
 577         }
 578
 579         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 580
 581         /* Send the ndis version */
 582         memset(init_packet, 0, sizeof(struct nvsp_message));
 583
 584         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 585                 ndis_version = 0x00060001;
 586         else
 587                 ndis_version = 0x0006001e;
 588
 589         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 590         init_packet->msg.v1_msg.
 591                 send_ndis_ver.ndis_major_ver =
 592                                 (ndis_version & 0xFFFF0000) >> 16;
 593         init_packet->msg.v1_msg.
 594                 send_ndis_ver.ndis_minor_ver =
 595                                 ndis_version & 0xFFFF;
 596
 597         trace_nvsp_send(ndev, init_packet);
 598
 599         /* Send the init request */
 600         ret = vmbus_sendpacket(device->channel, init_packet,
 601                                 sizeof(struct nvsp_message),
 602                                 (unsigned long)init_packet,
 603                                 VM_PKT_DATA_INBAND, 0);
 604         if (ret != 0)
 605                 goto cleanup;
 606
 607
 608         ret = netvsc_init_buf(device, net_device, device_info);
 609
 610 cleanup:
 611         return ret;
 612 }
 613
 614 /*
 615  * netvsc_device_remove - Callback when the root bus device is removed
 616  */
 617 void netvsc_device_remove(struct hv_device *device)
 618 {
 619         struct net_device *ndev = hv_get_drvdata(device);
 620         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 621         struct netvsc_device *net_device
 622                 = rtnl_dereference(net_device_ctx->nvdev);
 623         int i;
 624
 625         /*
 626          * Revoke receive buffer. If host is pre-Win2016 then tear down
 627          * receive buffer GPADL. Do the same for send buffer.
 628          */
 629         netvsc_revoke_recv_buf(device, net_device, ndev);
 630         if (vmbus_proto_version < VERSION_WIN10)
 631                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 632
 633         netvsc_revoke_send_buf(device, net_device, ndev);
 634         if (vmbus_proto_version < VERSION_WIN10)
 635                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 636
 637         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 638
 639         /* Disable NAPI and disassociate its context from the device. */
 640         for (i = 0; i < net_device->num_chn; i++) {
 641                 /* See also vmbus_reset_channel_cb(). */
 642                 napi_disable(&net_device->chan_table[i].napi);
 643                 netif_napi_del(&net_device->chan_table[i].napi);
 644         }
 645
 646         /*
 647          * At this point, no one should be accessing net_device
 648          * except in here
 649          */
 650         netdev_dbg(ndev, "net device safe to remove\n");
 651
 652         /* Now, we can close the channel safely */
 653         vmbus_close(device->channel);
 654
 655         /*
 656          * If host is Win2016 or higher then we do the GPADL tear down
 657          * here after VMBus is closed.
 658         */
 659         if (vmbus_proto_version >= VERSION_WIN10) {
 660                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 661                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 662         }
 663
 664         /* Release all resources */
 665         free_netvsc_device_rcu(net_device);
 666 }
 667
 668 #define RING_AVAIL_PERCENT_HIWATER 20
 669 #define RING_AVAIL_PERCENT_LOWATER 10
 670
 671 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 672                                          u32 index)
 673 {
 674         sync_change_bit(index, net_device->send_section_map);
 675 }
 676
 677 static void netvsc_send_tx_complete(struct net_device *ndev,
 678                                     struct netvsc_device *net_device,
 679                                     struct vmbus_channel *channel,
 680                                     const struct vmpacket_descriptor *desc,
 681                                     int budget)
 682 {
 683         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 684         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 685         u16 q_idx = 0;
 686         int queue_sends;
 687
 688         /* Notify the layer above us */
 689         if (likely(skb)) {
 690                 const struct hv_netvsc_packet *packet
 691                         = (struct hv_netvsc_packet *)skb->cb;
 692                 u32 send_index = packet->send_buf_index;
 693                 struct netvsc_stats *tx_stats;
 694
 695                 if (send_index != NETVSC_INVALID_INDEX)
 696                         netvsc_free_send_slot(net_device, send_index);
 697                 q_idx = packet->q_idx;
 698
 699                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 700
 701                 u64_stats_update_begin(&tx_stats->syncp);
 702                 tx_stats->packets += packet->total_packets;
 703                 tx_stats->bytes += packet->total_bytes;
 704                 u64_stats_update_end(&tx_stats->syncp);
 705
 706                 napi_consume_skb(skb, budget);
 707         }
 708
 709         queue_sends =
 710                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 711
 712         if (unlikely(net_device->destroy)) {
 713                 if (queue_sends == 0)
 714                         wake_up(&net_device->wait_drain);
 715         } else {
 716                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
 717
 718                 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
 719                     (hv_get_avail_to_write_percent(&channel->outbound) >
 720                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
 721                         netif_tx_wake_queue(txq);
 722                         ndev_ctx->eth_stats.wake_queue++;
 723                 }
 724         }
 725 }
 726
 727 static void netvsc_send_completion(struct net_device *ndev,
 728                                    struct netvsc_device *net_device,
 729                                    struct vmbus_channel *incoming_channel,
 730                                    const struct vmpacket_descriptor *desc,
 731                                    int budget)
 732 {
 733         const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 734
 735         switch (nvsp_packet->hdr.msg_type) {
 736         case NVSP_MSG_TYPE_INIT_COMPLETE:
 737         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 738         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 739         case NVSP_MSG5_TYPE_SUBCHANNEL:
 740                 /* Copy the response back */
 741                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 742                        sizeof(struct nvsp_message));
 743                 complete(&net_device->channel_init_wait);
 744                 break;
 745
 746         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 747                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
 748                                         desc, budget);
 749                 break;
 750
 751         default:
 752                 netdev_err(ndev,
 753                            "Unknown send completion type %d received!!\n",
 754                            nvsp_packet->hdr.msg_type);
 755         }
 756 }
 757
 758 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 759 {
 760         unsigned long *map_addr = net_device->send_section_map;
 761         unsigned int i;
 762
 763         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 764                 if (sync_test_and_set_bit(i, map_addr) == 0)
 765                         return i;
 766         }
 767
 768         return NETVSC_INVALID_INDEX;
 769 }
 770
 771 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 772                                     unsigned int section_index,
 773                                     u32 pend_size,
 774                                     struct hv_netvsc_packet *packet,
 775                                     struct rndis_message *rndis_msg,
 776                                     struct hv_page_buffer *pb,
 777                                     bool xmit_more)
 778 {
 779         char *start = net_device->send_buf;
 780         char *dest = start + (section_index * net_device->send_section_size)
 781                      + pend_size;
 782         int i;
 783         u32 padding = 0;
 784         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 785                 packet->page_buf_cnt;
 786         u32 remain;
 787
 788         /* Add padding */
 789         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 790         if (xmit_more && remain) {
 791                 padding = net_device->pkt_align - remain;
 792                 rndis_msg->msg_len += padding;
 793                 packet->total_data_buflen += padding;
 794         }
 795
 796         for (i = 0; i < page_count; i++) {
 797                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
 798                 u32 offset = pb[i].offset;
 799                 u32 len = pb[i].len;
 800
 801                 memcpy(dest, (src + offset), len);
 802                 dest += len;
 803         }
 804
 805         if (padding)
 806                 memset(dest, 0, padding);
 807 }
 808
 809 static inline int netvsc_send_pkt(
 810         struct hv_device *device,
 811         struct hv_netvsc_packet *packet,
 812         struct netvsc_device *net_device,
 813         struct hv_page_buffer *pb,
 814         struct sk_buff *skb)
 815 {
 816         struct nvsp_message nvmsg;
 817         struct nvsp_1_message_send_rndis_packet *rpkt =
 818                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 819         struct netvsc_channel * const nvchan =
 820                 &net_device->chan_table[packet->q_idx];
 821         struct vmbus_channel *out_channel = nvchan->channel;
 822         struct net_device *ndev = hv_get_drvdata(device);
 823         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 824         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 825         u64 req_id;
 826         int ret;
 827         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
 828
 829         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 830         if (skb)
 831                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 832         else
 833                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 834
 835         rpkt->send_buf_section_index = packet->send_buf_index;
 836         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 837                 rpkt->send_buf_section_size = 0;
 838         else
 839                 rpkt->send_buf_section_size = packet->total_data_buflen;
 840
 841         req_id = (ulong)skb;
 842
 843         if (out_channel->rescind)
 844                 return -ENODEV;
 845
 846         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
 847
 848         if (packet->page_buf_cnt) {
 849                 if (packet->cp_partial)
 850                         pb += packet->rmsg_pgcnt;
 851
 852                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 853                                                   pb, packet->page_buf_cnt,
 854                                                   &nvmsg, sizeof(nvmsg),
 855                                                   req_id);
 856         } else {
 857                 ret = vmbus_sendpacket(out_channel,
 858                                        &nvmsg, sizeof(nvmsg),
 859                                        req_id, VM_PKT_DATA_INBAND,
 860                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 861         }
 862
 863         if (ret == 0) {
 864                 atomic_inc_return(&nvchan->queue_sends);
 865
 866                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 867                         netif_tx_stop_queue(txq);
 868                         ndev_ctx->eth_stats.stop_queue++;
 869                 }
 870         } else if (ret == -EAGAIN) {
 871                 netif_tx_stop_queue(txq);
 872                 ndev_ctx->eth_stats.stop_queue++;
 873         } else {
 874                 netdev_err(ndev,
 875                            "Unable to send packet pages %u len %u, ret %d\n",
 876                            packet->page_buf_cnt, packet->total_data_buflen,
 877                            ret);
 878         }
 879
 880         if (netif_tx_queue_stopped(txq) &&
 881             atomic_read(&nvchan->queue_sends) < 1 &&
 882             !net_device->tx_disable) {
 883                 netif_tx_wake_queue(txq);
 884                 ndev_ctx->eth_stats.wake_queue++;
 885                 if (ret == -EAGAIN)
 886                         ret = -ENOSPC;
 887         }
 888
 889         return ret;
 890 }
 891
 892 /* Move packet out of multi send data (msd), and clear msd */
 893 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 894                                 struct sk_buff **msd_skb,
 895                                 struct multi_send_data *msdp)
 896 {
 897         *msd_skb = msdp->skb;
 898         *msd_send = msdp->pkt;
 899         msdp->skb = NULL;
 900         msdp->pkt = NULL;
 901         msdp->count = 0;
 902 }
 903
 904 /* RCU already held by caller */
 905 int netvsc_send(struct net_device *ndev,
 906                 struct hv_netvsc_packet *packet,
 907                 struct rndis_message *rndis_msg,
 908                 struct hv_page_buffer *pb,
 909                 struct sk_buff *skb,
 910                 bool xdp_tx)
 911 {
 912         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 913         struct netvsc_device *net_device
 914                 = rcu_dereference_bh(ndev_ctx->nvdev);
 915         struct hv_device *device = ndev_ctx->device_ctx;
 916         int ret = 0;
 917         struct netvsc_channel *nvchan;
 918         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 919         unsigned int section_index = NETVSC_INVALID_INDEX;
 920         struct multi_send_data *msdp;
 921         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 922         struct sk_buff *msd_skb = NULL;
 923         bool try_batch, xmit_more;
 924
 925         /* If device is rescinded, return error and packet will get dropped. */
 926         if (unlikely(!net_device || net_device->destroy))
 927                 return -ENODEV;
 928
 929         nvchan = &net_device->chan_table[packet->q_idx];
 930         packet->send_buf_index = NETVSC_INVALID_INDEX;
 931         packet->cp_partial = false;
 932
 933         /* Send a control message or XDP packet directly without accessing
 934          * msd (Multi-Send Data) field which may be changed during data packet
 935          * processing.
 936          */
 937         if (!skb || xdp_tx)
 938                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
 939
 940         /* batch packets in send buffer if possible */
 941         msdp = &nvchan->msd;
 942         if (msdp->pkt)
 943                 msd_len = msdp->pkt->total_data_buflen;
 944
 945         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 946         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 947             net_device->send_section_size) {
 948                 section_index = msdp->pkt->send_buf_index;
 949
 950         } else if (try_batch && msd_len + packet->rmsg_size <
 951                    net_device->send_section_size) {
 952                 section_index = msdp->pkt->send_buf_index;
 953                 packet->cp_partial = true;
 954
 955         } else if (pktlen + net_device->pkt_align <
 956                    net_device->send_section_size) {
 957                 section_index = netvsc_get_next_send_section(net_device);
 958                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
 959                         ++ndev_ctx->eth_stats.tx_send_full;
 960                 } else {
 961                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 962                         msd_len = 0;
 963                 }
 964         }
 965
 966         /* Keep aggregating only if stack says more data is coming
 967          * and not doing mixed modes send and not flow blocked
 968          */
 969         xmit_more = netdev_xmit_more() &&
 970                 !packet->cp_partial &&
 971                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
 972
 973         if (section_index != NETVSC_INVALID_INDEX) {
 974                 netvsc_copy_to_send_buf(net_device,
 975                                         section_index, msd_len,
 976                                         packet, rndis_msg, pb, xmit_more);
 977
 978                 packet->send_buf_index = section_index;
 979
 980                 if (packet->cp_partial) {
 981                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 982                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 983                 } else {
 984                         packet->page_buf_cnt = 0;
 985                         packet->total_data_buflen += msd_len;
 986                 }
 987
 988                 if (msdp->pkt) {
 989                         packet->total_packets += msdp->pkt->total_packets;
 990                         packet->total_bytes += msdp->pkt->total_bytes;
 991                 }
 992
 993                 if (msdp->skb)
 994                         dev_consume_skb_any(msdp->skb);
 995
 996                 if (xmit_more) {
 997                         msdp->skb = skb;
 998                         msdp->pkt = packet;
 999                         msdp->count++;
1000                 } else {
1001                         cur_send = packet;
1002                         msdp->skb = NULL;
1003                         msdp->pkt = NULL;
1004                         msdp->count = 0;
1005                 }
1006         } else {
1007                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1008                 cur_send = packet;
1009         }
1010
1011         if (msd_send) {
1012                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1013                                             NULL, msd_skb);
1014
1015                 if (m_ret != 0) {
1016                         netvsc_free_send_slot(net_device,
1017                                               msd_send->send_buf_index);
1018                         dev_kfree_skb_any(msd_skb);
1019                 }
1020         }
1021
1022         if (cur_send)
1023                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1024
1025         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1026                 netvsc_free_send_slot(net_device, section_index);
1027
1028         return ret;
1029 }
1030
1031 /* Send pending recv completions */
1032 static int send_recv_completions(struct net_device *ndev,
1033                                  struct netvsc_device *nvdev,
1034                                  struct netvsc_channel *nvchan)
1035 {
1036         struct multi_recv_comp *mrc = &nvchan->mrc;
1037         struct recv_comp_msg {
1038                 struct nvsp_message_header hdr;
1039                 u32 status;
1040         }  __packed;
1041         struct recv_comp_msg msg = {
1042                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1043         };
1044         int ret;
1045
1046         while (mrc->first != mrc->next) {
1047                 const struct recv_comp_data *rcd
1048                         = mrc->slots + mrc->first;
1049
1050                 msg.status = rcd->status;
1051                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1052                                        rcd->tid, VM_PKT_COMP, 0);
1053                 if (unlikely(ret)) {
1054                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1055
1056                         ++ndev_ctx->eth_stats.rx_comp_busy;
1057                         return ret;
1058                 }
1059
1060                 if (++mrc->first == nvdev->recv_completion_cnt)
1061                         mrc->first = 0;
1062         }
1063
1064         /* receive completion ring has been emptied */
1065         if (unlikely(nvdev->destroy))
1066                 wake_up(&nvdev->wait_drain);
1067
1068         return 0;
1069 }
1070
1071 /* Count how many receive completions are outstanding */
1072 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1073                                  const struct multi_recv_comp *mrc,
1074                                  u32 *filled, u32 *avail)
1075 {
1076         u32 count = nvdev->recv_completion_cnt;
1077
1078         if (mrc->next >= mrc->first)
1079                 *filled = mrc->next - mrc->first;
1080         else
1081                 *filled = (count - mrc->first) + mrc->next;
1082
1083         *avail = count - *filled - 1;
1084 }
1085
1086 /* Add receive complete to ring to send to host. */
1087 static void enq_receive_complete(struct net_device *ndev,
1088                                  struct netvsc_device *nvdev, u16 q_idx,
1089                                  u64 tid, u32 status)
1090 {
1091         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1092         struct multi_recv_comp *mrc = &nvchan->mrc;
1093         struct recv_comp_data *rcd;
1094         u32 filled, avail;
1095
1096         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1097
1098         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1099                 send_recv_completions(ndev, nvdev, nvchan);
1100                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1101         }
1102
1103         if (unlikely(!avail)) {
1104                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1105                            q_idx, tid);
1106                 return;
1107         }
1108
1109         rcd = mrc->slots + mrc->next;
1110         rcd->tid = tid;
1111         rcd->status = status;
1112
1113         if (++mrc->next == nvdev->recv_completion_cnt)
1114                 mrc->next = 0;
1115 }
1116
1117 static int netvsc_receive(struct net_device *ndev,
1118                           struct netvsc_device *net_device,
1119                           struct netvsc_channel *nvchan,
1120                           const struct vmpacket_descriptor *desc,
1121                           const struct nvsp_message *nvsp)
1122 {
1123         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1124         struct vmbus_channel *channel = nvchan->channel;
1125         const struct vmtransfer_page_packet_header *vmxferpage_packet
1126                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1127         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1128         char *recv_buf = net_device->recv_buf;
1129         u32 status = NVSP_STAT_SUCCESS;
1130         int i;
1131         int count = 0;
1132
1133         /* Make sure this is a valid nvsp packet */
1134         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1135                 netif_err(net_device_ctx, rx_err, ndev,
1136                           "Unknown nvsp packet type received %u\n",
1137                           nvsp->hdr.msg_type);
1138                 return 0;
1139         }
1140
1141         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1142                 netif_err(net_device_ctx, rx_err, ndev,
1143                           "Invalid xfer page set id - expecting %x got %x\n",
1144                           NETVSC_RECEIVE_BUFFER_ID,
1145                           vmxferpage_packet->xfer_pageset_id);
1146                 return 0;
1147         }
1148
1149         count = vmxferpage_packet->range_cnt;
1150
1151         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1152         for (i = 0; i < count; i++) {
1153                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1154                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1155                 void *data;
1156                 int ret;
1157
1158                 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1159                         nvchan->rsc.cnt = 0;
1160                         status = NVSP_STAT_FAIL;
1161                         netif_err(net_device_ctx, rx_err, ndev,
1162                                   "Packet offset:%u + len:%u too big\n",
1163                                   offset, buflen);
1164
1165                         continue;
1166                 }
1167
1168                 data = recv_buf + offset;
1169
1170                 nvchan->rsc.is_last = (i == count - 1);
1171
1172                 trace_rndis_recv(ndev, q_idx, data);
1173
1174                 /* Pass it to the upper layer */
1175                 ret = rndis_filter_receive(ndev, net_device,
1176                                            nvchan, data, buflen);
1177
1178                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1179                         status = NVSP_STAT_FAIL;
1180         }
1181
1182         enq_receive_complete(ndev, net_device, q_idx,
1183                              vmxferpage_packet->d.trans_id, status);
1184
1185         return count;
1186 }
1187
1188 static void netvsc_send_table(struct net_device *ndev,
1189                               struct netvsc_device *nvscdev,
1190                               const struct nvsp_message *nvmsg,
1191                               u32 msglen)
1192 {
1193         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1194         u32 count, offset, *tab;
1195         int i;
1196
1197         count = nvmsg->msg.v5_msg.send_table.count;
1198         offset = nvmsg->msg.v5_msg.send_table.offset;
1199
1200         if (count != VRSS_SEND_TAB_SIZE) {
1201                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1202                 return;
1203         }
1204
1205         /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1206          * wrong due to a host bug. So fix the offset here.
1207          */
1208         if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1209             msglen >= sizeof(struct nvsp_message_header) +
1210             sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1211                 offset = sizeof(struct nvsp_message_header) +
1212                          sizeof(union nvsp_6_message_uber);
1213
1214         /* Boundary check for all versions */
1215         if (offset > msglen - count * sizeof(u32)) {
1216                 netdev_err(ndev, "Received send-table offset too big:%u\n",
1217                            offset);
1218                 return;
1219         }
1220
1221         tab = (void *)nvmsg + offset;
1222
1223         for (i = 0; i < count; i++)
1224                 net_device_ctx->tx_table[i] = tab[i];
1225 }
1226
1227 static void netvsc_send_vf(struct net_device *ndev,
1228                            const struct nvsp_message *nvmsg)
1229 {
1230         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1231
1232         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1233         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1234         netdev_info(ndev, "VF slot %u %s\n",
1235                     net_device_ctx->vf_serial,
1236                     net_device_ctx->vf_alloc ? "added" : "removed");
1237 }
1238
1239 static void netvsc_receive_inband(struct net_device *ndev,
1240                                   struct netvsc_device *nvscdev,
1241                                   const struct nvsp_message *nvmsg,
1242                                   u32 msglen)
1243 {
1244         switch (nvmsg->hdr.msg_type) {
1245         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1246                 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1247                 break;
1248
1249         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1250                 netvsc_send_vf(ndev, nvmsg);
1251                 break;
1252         }
1253 }
1254
1255 static int netvsc_process_raw_pkt(struct hv_device *device,
1256                                   struct netvsc_channel *nvchan,
1257                                   struct netvsc_device *net_device,
1258                                   struct net_device *ndev,
1259                                   const struct vmpacket_descriptor *desc,
1260                                   int budget)
1261 {
1262         struct vmbus_channel *channel = nvchan->channel;
1263         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1264         u32 msglen = hv_pkt_datalen(desc);
1265
1266         trace_nvsp_recv(ndev, channel, nvmsg);
1267
1268         switch (desc->type) {
1269         case VM_PKT_COMP:
1270                 netvsc_send_completion(ndev, net_device, channel,
1271                                        desc, budget);
1272                 break;
1273
1274         case VM_PKT_DATA_USING_XFER_PAGES:
1275                 return netvsc_receive(ndev, net_device, nvchan,
1276                                       desc, nvmsg);
1277                 break;
1278
1279         case VM_PKT_DATA_INBAND:
1280                 netvsc_receive_inband(ndev, net_device, nvmsg, msglen);
1281                 break;
1282
1283         default:
1284                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1285                            desc->type, desc->trans_id);
1286                 break;
1287         }
1288
1289         return 0;
1290 }
1291
1292 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1293 {
1294         struct vmbus_channel *primary = channel->primary_channel;
1295
1296         return primary ? primary->device_obj : channel->device_obj;
1297 }
1298
1299 /* Network processing softirq
1300  * Process data in incoming ring buffer from host
1301  * Stops when ring is empty or budget is met or exceeded.
1302  */
1303 int netvsc_poll(struct napi_struct *napi, int budget)
1304 {
1305         struct netvsc_channel *nvchan
1306                 = container_of(napi, struct netvsc_channel, napi);
1307         struct netvsc_device *net_device = nvchan->net_device;
1308         struct vmbus_channel *channel = nvchan->channel;
1309         struct hv_device *device = netvsc_channel_to_device(channel);
1310         struct net_device *ndev = hv_get_drvdata(device);
1311         int work_done = 0;
1312         int ret;
1313
1314         /* If starting a new interval */
1315         if (!nvchan->desc)
1316                 nvchan->desc = hv_pkt_iter_first(channel);
1317
1318         while (nvchan->desc && work_done < budget) {
1319                 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1320                                                     ndev, nvchan->desc, budget);
1321                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1322         }
1323
1324         /* Send any pending receive completions */
1325         ret = send_recv_completions(ndev, net_device, nvchan);
1326
1327         /* If it did not exhaust NAPI budget this time
1328          *  and not doing busy poll
1329          * then re-enable host interrupts
1330          *  and reschedule if ring is not empty
1331          *   or sending receive completion failed.
1332          */
1333         if (work_done < budget &&
1334             napi_complete_done(napi, work_done) &&
1335             (ret || hv_end_read(&channel->inbound)) &&
1336             napi_schedule_prep(napi)) {
1337                 hv_begin_read(&channel->inbound);
1338                 __napi_schedule(napi);
1339         }
1340
1341         /* Driver may overshoot since multiple packets per descriptor */
1342         return min(work_done, budget);
1343 }
1344
1345 /* Call back when data is available in host ring buffer.
1346  * Processing is deferred until network softirq (NAPI)
1347  */
1348 void netvsc_channel_cb(void *context)
1349 {
1350         struct netvsc_channel *nvchan = context;
1351         struct vmbus_channel *channel = nvchan->channel;
1352         struct hv_ring_buffer_info *rbi = &channel->inbound;
1353
1354         /* preload first vmpacket descriptor */
1355         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1356
1357         if (napi_schedule_prep(&nvchan->napi)) {
1358                 /* disable interrupts from host */
1359                 hv_begin_read(rbi);
1360
1361                 __napi_schedule_irqoff(&nvchan->napi);
1362         }
1363 }
1364
1365 /*
1366  * netvsc_device_add - Callback when the device belonging to this
1367  * driver is added
1368  */
1369 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1370                                 const struct netvsc_device_info *device_info)
1371 {
1372         int i, ret = 0;
1373         struct netvsc_device *net_device;
1374         struct net_device *ndev = hv_get_drvdata(device);
1375         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1376
1377         net_device = alloc_net_device();
1378         if (!net_device)
1379                 return ERR_PTR(-ENOMEM);
1380
1381         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1382                 net_device_ctx->tx_table[i] = 0;
1383
1384         /* Because the device uses NAPI, all the interrupt batching and
1385          * control is done via Net softirq, not the channel handling
1386          */
1387         set_channel_read_mode(device->channel, HV_CALL_ISR);
1388
1389         /* If we're reopening the device we may have multiple queues, fill the
1390          * chn_table with the default channel to use it before subchannels are
1391          * opened.
1392          * Initialize the channel state before we open;
1393          * we can be interrupted as soon as we open the channel.
1394          */
1395
1396         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1397                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1398
1399                 nvchan->channel = device->channel;
1400                 nvchan->net_device = net_device;
1401                 u64_stats_init(&nvchan->tx_stats.syncp);
1402                 u64_stats_init(&nvchan->rx_stats.syncp);
1403
1404                 ret = xdp_rxq_info_reg(&nvchan->xdp_rxq, ndev, i);
1405
1406                 if (ret) {
1407                         netdev_err(ndev, "xdp_rxq_info_reg fail: %d\n", ret);
1408                         goto cleanup2;
1409                 }
1410
1411                 ret = xdp_rxq_info_reg_mem_model(&nvchan->xdp_rxq,
1412                                                  MEM_TYPE_PAGE_SHARED, NULL);
1413
1414                 if (ret) {
1415                         netdev_err(ndev, "xdp reg_mem_model fail: %d\n", ret);
1416                         goto cleanup2;
1417                 }
1418         }
1419
1420         /* Enable NAPI handler before init callbacks */
1421         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1422                        netvsc_poll, NAPI_POLL_WEIGHT);
1423
1424         /* Open the channel */
1425         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1426                          netvsc_ring_bytes,  NULL, 0,
1427                          netvsc_channel_cb, net_device->chan_table);
1428
1429         if (ret != 0) {
1430                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1431                 goto cleanup;
1432         }
1433
1434         /* Channel is opened */
1435         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1436
1437         napi_enable(&net_device->chan_table[0].napi);
1438
1439         /* Connect with the NetVsp */
1440         ret = netvsc_connect_vsp(device, net_device, device_info);
1441         if (ret != 0) {
1442                 netdev_err(ndev,
1443                         "unable to connect to NetVSP - %d\n", ret);
1444                 goto close;
1445         }
1446
1447         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1448          * populated.
1449          */
1450         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1451
1452         return net_device;
1453
1454 close:
1455         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1456         napi_disable(&net_device->chan_table[0].napi);
1457
1458         /* Now, we can close the channel safely */
1459         vmbus_close(device->channel);
1460
1461 cleanup:
1462         netif_napi_del(&net_device->chan_table[0].napi);
1463
1464 cleanup2:
1465         free_netvsc_device(&net_device->rcu);
1466
1467         return ERR_PTR(ret);
1468 }